Alerts: Alerts Triggering Because of lack of data

[WilliamBr6]

We have a dataset that we recently added to our connection. Sometimes it will have large spikes that we need to be aware of, so I set up an alert for anomalies outside the 95% interval. However, this alert has been triggering every single day, saying the daily value was 400,000% more than expected, despite being an expected number. 

I believe this is because 10/24 was the first day we started ingesting data, and it is assessing every day before that as 0 rows when determining what is an anomaly and what isn't. If this is the case, is there a set number of days that the alert system looks back when assessing anomalies? Is there a way to change this to the first day we started ingesting data for this dataset? 

Thanks!

[MandyGeorge]

There isn't a set number of days for the alert look back, at least not that Adobe has shared with users.

It is possible that over the next week or so it should pick up the fact that the data is an expected value. Unfortunaly it isn't recommended to use date ranges in alerts, otherwise that would be my suggestion.

Maybe you could add a condition to the alert to only fire if there is greater than X hits or less than Y hits? For example if you're expecting around 1000 hits a day, set up conditions to only fire if there is greater than 1200 or less than 800.

Even if it's just temporary until it builds up some historical data so that it doesn't fire every day.

If this persists, you could always reach out to customer support, see if there is anything they can do to help.

[Jennifer_Dungan]

Given the newness of the metric, the anomaly detection is going to be wonky up front... you might want to create your own threshold value, rather than relying on the automatic calculations... 

As Mandy said, you could look at "Greater than X hits per day/hour/etc", or you can create a calculated metric based on the number of visits... (assuming the inflation you are seeing isn't related to a correlating Visit spike).

I have had similar, very extreme, anomaly percentages on new values before... while we don't know the exact lookback that Adobe is using, I would guess it wouldn't be shorter than a few weeks, and likely a few months...